Three kinds of break in Banco de Moçambique’s July 2026 pack, and how to chain around them

July 31, 2026

Wordless fine-line illustration: A railway engineer joining two lengths of track on a river bridge where the sleepers on one side sit a few centimetres lower than the other, with a surveyor's level resting on the rail

Opening balances that miss the last close, quarters that vanish from a table and GDP bars that move: a longitudinal analyst’s checklist.

The July 2026 edition of Banco de Moçambique’s statistical summary, posted on 6 August, opens its June reserves bridge at US$3,503.80m. The previous edition closed May at US$3,496.6m. The US$7.2m difference is small, but it is the third time in six hand-offs between consecutive editions this year that a closing balance has not become the next opening balance.

It is also one of three different kinds of discontinuity in the same document. The second is a table whose history was cut between editions, and the third is a GDP chart whose 2025 bars have moved. Each breaks a longitudinal series in a different way, and each needs a different repair.

Break one: the bridge does not chain

Three of six hand-offs fail, and each time the new opening agrees with the monetary table instead.

The January edition closed December at US$4,086.4m, and the February edition opened January at US$4,123.1m, a difference of US$36.7m. The April edition closed March at US$3,471.1m, and the May edition opened April at what is evidently a typing slip for US$3,486.4m, a difference of US$15.3m once the slip is corrected. The June edition closed May at US$3,496.6m and the July edition opens June at US$3,503.80m, a difference of US$7.2m. The other three hand-offs tie exactly: US$4,151.9m from the February to the March edition, US$4,258.1m from March to April, and US$3,470.1m from May to June.

In every failing case the new opening equals the monthly reserves figure in Quadro 1, and in every case the bridge’s own closing sat below that figure. The gaps shrink from US$36.7m to US$15.2m to US$7.2m. The mechanism is that the bridge is re-based to the stock series at the start of each month, so differences between the bridge’s accumulated flows and the stock are written off silently. The stock is the series, and the bridge is a commentary.

Break two: the quarters that left the table

The 2025 quarterly columns in the trade and investment tables disappeared after the May edition.

From the February to the May editions, the goods trade and direct investment tables showed each quarter of 2025 separately, next to the annual total. From the June edition, and again in July, they show only 2023, 2024, 2025 and the first quarter of 2026. For the new quarter that means goods exports of US$1,853.0m, goods imports of US$1,941.7m and direct investment of US$1,286.0m. A reader who wants to set the first quarter of 2026 against the first quarter of 2025, when investment stood at US$1,626.3m, cannot do it from the July file alone.

The quarterly history has not been lost. The Bank’s Boletim Estatístico for June still carries separate 2025 quarters in its balance-of-payments tables, with direct investment of US$2,195.2m in the third quarter and US$969.0m in the fourth. But the two publications now answer different questions, and an analyst who uses the monthly pack as a stand-alone source has to archive the older editions to keep the history.

The practical cost falls on small teams. A bank’s research desk or a development-finance analyst who relied on the monthly pack to refresh a quarterly dashboard now needs to join two publications, and to check that quarters in one agree with quarters in the other before any chart is published.

Break three: a GDP chart that restates 2025

The bar heights for the same quarters differ between editions, with no note.

Figura 6 in each edition charts the growth of real GDP by quarter. In the March edition the four quarters of 2025 are labelled minus 3.92, minus 0.94, minus 0.85 and plus 4.67. In July they are minus 2.9, minus 1.1, minus 0.6 and plus 5.1, with plus 0.1 added for the first quarter of 2026. INE reported the 0.1% rise for the first quarter against a 2.9% contraction a year earlier, as AIM reported on 18 June, so the July chart reflects a restatement at the source.

The figures are a chart image, so no parser can read them, and the footnote says nothing about revision. A user who copied the March bars into a model has the wrong history, and cannot know it without opening both PDFs. The same edition shows how far national and capital-city price series have diverged: INE’s annual inflation for June is 7.51% nationally and 4.16% for the city of Maputo, according to Quadro 7, and Jornal Económico reported the national rate on 13 July. Different geography is a different series, and a pricing model built on the capital’s index will under-read the national pass-through of the fuel shock that began in May.

A chaining discipline

Pick the governing table, freeze vintages and rebuild history from the richer publication.

For each series, name one governing table and record it in the model’s documentation: Quadro 1 for reserves, the Boletim for quarterly balance-of-payments history, INE’s own releases for GDP. Keep every monthly PDF in a dated archive, hash it, and store revisions as separate vintages instead of overwriting earlier values. Re-base on the governing table whenever a secondary table disagrees, and log the size of the gap, since a widening gap is itself information.

For funds and lenders that benchmark Mozambican risk on these series, the cheapest improvement is a short methodological request to the Departamento de Estatística e Reporte: a standing revisions note in each monthly pack, listing any restated quarter and any table whose history was shortened. The Bank has the information, and the request costs one paragraph.

A last habit is worth adopting for any decision that rests on a single month: write down, beside the number, the edition it came from and the table that governs it. When the figure changes in a later vintage, the note turns a puzzling discrepancy into a ten-minute reconciliation.